Deep Time Series Generation Using Domain Knowledge

• Applicant: Professorin Dr. Sophie Fellenz
• Subject Area: Image and Language Processing, Computer Graphics and Visualisation, Human Computer Interaction, Ubiquitous and Wearable Computing
• Term: since 2022
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 459419731

Data in chemical process engineering (CPE) is scarce and expensive to produce. Recent machine learning (ML) methods can generate high-quality image and text data. In the time series domain, where comparatively little data is available for training, generating realistic artificial data is difficult. It would be vital to generate data in these cases as this would enable the development of methods that rely on larger datasets. To do this, we need to provide additional information for network training in the form of external knowledge. Knowledge may come in many different forms. In the domain of chemical process engineering, Projects B1 and B2 will provide us with the following domain knowledge: the structure of the chemical plants, the properties of chemical compounds that are processed, and mathematical equations that describe the physical properties, boundary conditions, and the development of a system over time through differential equations. This project aims to integrate different forms of domain knowledge with machine learning techniques to generate realistic time series data for chemical process engineering. Projects A1 and A3 will use this additional data for the development of machine learning methods for time series data, and Project A2 will verify our methods. Additionally, the controlled data generation will enable Project B1 and B2 to compare model predictions with potential experimental outcomes.

• DFG Programme: Research Units
• Subproject of: FOR 5359: KI-FOR: Deep Learning on Sparse Chemical Process Data